[PATCH] update the man pages for the prefork process model
garming at catalyst.net.nz
Wed Jan 30 21:08:07 UTC 2019
On 30/01/19 5:44 PM, Douglas Bagnall via samba-technical wrote:
> On 29/01/19 12:07 PM, Tim Beale via samba-technical wrote:
>> The samba man pages lack detail on how to enable the prefork process
>> model. It seems that Gary posted a patch to address this some time ago,
>> but this never ended up landing in master.
>> I've rebased his patch and tweaked the content ever so slightly. Review
> LGTM, but it raises again the question:
> why do people have to enable prefork, when it is almost always the best
> option? (AFAICT, from my sheltered position).
There were a few reasons before 4.10 coming up, like NETLOGON (although
it always had the downside of clients holding connections, so that was
probably degenerate enough to begin with). But now it's probably better
in nearly every way -- it now also includes a multi-process KDC. Perhaps
the forking mode in standard helps in some extremely concurrent cases,
but you need a lot of RAM + the connections probably have to have a
particular usage pattern to really make any difference in throughput.
Frankly, setting up prefork with a huge amount of active processes would
do just as well, be more reliable/stable and eliminate the extra fork time.
> Why not make it the default?
There's also the issue of what does 'standard' mean -- because it's kind
of a synonym for default. Doesn't it also make sense to modify standard
to prefork and introduce a new option for forking instead?
More information about the samba-technical